Aerospace and Mechanical Engineering

1st Exam of Computational Fluid Dynamics

January 24th, 2025, 8:00 AM
Room: IA
Closed Book Exam
Duration: 2 hours
Name: ________________________________________________________________________ Number: ______________

We are all expected to abide by the IST standards for academic excellence and integrity.

Problem 1 (0.75+0.75+0.75+1.25+1.0+1.0+0.5+1.0+0.5 = 7.5 Marks)

Consider the one dimensional bar, below, where a thermal conduction process takes place,
subjected to a uniform thermal source.
𝜕𝑇
In terms of boundary conditions, consider an adiabatic condition on the left (𝑥 = 0 → 𝜕𝑥 = 0) and
an imposed value on the right (𝑥 = 1 → 𝑇 = 100).
The RED points refer to the Finite Differences Domain and the BLACK points refer to the
centroids of the Control Volumes defined by the Black rectangular 1D Cells.

Under this conditions the ODE governing the process is:

∂𝑇 ∂𝑇 𝜕2 𝑇
= ∇ ∙ (∇𝑇) + 6 → = +6
∂𝑡 ∂t 𝜕𝑥2
Regarding the Finite Differences approach:

a) Discretize the equation at point i with a second order central scheme in space and first
order explicit Euler in time. Then fill the row 4 of the system of equations below.

𝜕𝑇 𝜕 2 𝑇
= +6
𝜕𝑡 𝜕𝑥 2
Applying the classic and well known schemes, as asked:

𝑇𝑖𝑛+1 − 𝑇𝑖𝑛 𝑇𝑖+1 𝑛

− 2𝑇𝑖𝑛 + 𝑇𝑖−1
𝑛
0.5
= +6
𝛥𝑡 (Δ𝑥)2
Simplifying and rearranging in order to fill the fourth row:
𝑛
𝑇𝑖+1 − 2𝑇𝑖𝑛 + 𝑇𝑖−1
𝑛
𝑇𝑖𝑛+1 = 𝑇𝑖𝑛 + 𝛥𝑡 + 6𝛥𝑡
(Δ𝑥)2
𝛥𝑡 𝛥𝑡 𝛥𝑡
𝑇𝑖𝑛+1 = (1 − 2 ) 𝑇𝑖𝑛 + 𝑛
𝑇𝑖+1 + 𝑇 𝑛 + 6𝛥𝑡
(Δ𝑥) 2 (Δ𝑥)2 (Δ𝑥)2 𝑖−1
𝛥𝑡 𝛥𝑡 𝛥𝑡
𝑇4𝑛+1 = (1 − 2 ) 𝑇4𝑛 + 𝑇5𝑛 + 𝑇 𝑛 + 6𝛥𝑡
(Δ𝑥) 2 (Δ𝑥) 2 (Δ𝑥)2 3
Filling the row as below 0.25

1
 𝑻𝟏

𝑻𝟐

𝑻𝟑 =
𝛥𝑡 𝛥𝑡 𝛥𝑡
1 𝑻𝒏+𝟏
𝟒 (1 − 2
(Δ𝑥) 2
) 𝑇4𝑛 +
(Δ𝑥) 2
𝑇5𝑛 + 𝑇 𝑛 + 6𝛥𝑡
(Δ𝑥)2 3

𝑻𝟓

𝑻𝟔

𝑻𝟕

b) Discretize the equation at point i with a second order central scheme in space and first
order Implicit Euler in time. Then fill the row 4 of the system of equations below.

The same as before, just changing the time instant in the spatial derivative:
𝑇𝑖𝑛+1 − 𝑇𝑖𝑛 𝑇𝑖+1𝑛+1
− 2𝑇𝑖𝑛+1 + 𝑇𝑖−1
𝑛+1
= +6 0.25
𝛥𝑡 (Δ𝑥)2
Simplifying and rearranging in order to fill the fourth row:
𝑛+1
𝑇𝑖+1 − 2𝑇𝑖𝑛+1 + 𝑇𝑖−1
𝑛+1
𝑇𝑖𝑛+1 − 𝛥𝑡 = 𝑇𝑖𝑛 + +6𝛥𝑡
(Δ𝑥)2
𝛥𝑡 𝑛+1 𝛥𝑡 𝛥𝑡
− 𝑇𝑖−1 + (1 + 2 ) 𝑇𝑖𝑛+1 − 𝑇 𝑛+1 = 𝑇𝑖𝑛 + 6𝛥𝑡
(Δ𝑥)2 (Δ𝑥) 2 (Δ𝑥)2 𝑖+1
𝛥𝑡 𝛥𝑡 𝛥𝑡
− 𝑇3
𝑛+1
+ (1 + 2 ) 𝑇4
𝑛+1
− 𝑇 𝑛+1 = 𝑇4𝑛 + 6𝛥𝑡
(Δ𝑥)2 (Δ𝑥)2 (Δ𝑥)2 5
Filling the row as below 0.5

𝑻𝟏

𝑻𝟐

𝑻𝒏+𝟏
𝟑 =
𝛥𝑡 𝛥𝑡 𝛥𝑡
−
(Δ𝑥)2
1+2
(Δ𝑥)2
−
(Δ𝑥)2 𝑻𝒏+𝟏
𝟒 𝑇4𝑛 + 6𝛥𝑡

𝑻𝒏+𝟏
𝟓

𝑻𝟔

𝑻𝟕

2
 c) If the expected result to be attained was a steady state thermal field, fully explain the
different procedures to be considered, under the different numerical approaches
explored in the previous items: Explicit vs Implicit.
1 - Generally speaking, both methods, explicit or implicit, will reach or approach the steady
state after some physical time, following the usual time stepping. The main difference is
that the implicit method is much more expensive per time step, but allows for a much larger 0.15
time step than the explicit method. In the end of the day, the time resolution is not an issue
since the goal is only the final state.
2 - For the Explicit method, simply advance in time with the largest possible time step, only 0.3
limited by some stability constrains.
3 - For the Implicit method, simply remove the time derivative, which means Δ𝑡 = ∞.
(looking to the system in the previous item, all terms should be divided by the time step) 0.3
Then solve the previous system of equations to obtain the steady state solution. The matrix
on the LHS is diagonally dominant and thus a solution is expected to be ‘easily’ obtained.

d) Consider now, only for this question, the PDE with no source term:
∂T 𝜕2 𝑇
=
∂t 𝜕𝑥2
Apply the von Neumann stability analysis for one of the previous items of your choice,
item a) or item b), and find the stability limit. Consider the parameter to be constrained,
Δ𝑡
the diffusive criterion, as 𝛾 = (Δ𝑥)2 .
Only the Explicit method will be presented with detail:
Starting with the FD equation:
𝑇𝑖𝑛+1 − 𝑇𝑖𝑛 = 𝛾(𝑇𝑖+1
𝑛
− 2𝑇𝑖𝑛 + 𝑇𝑖−1
𝑛
)
Applying the Fourier mode:
𝐴𝑛+1 𝑒𝐼𝜃𝑖 − 𝐴𝑛 𝑒𝐼𝜃𝑖 = 𝛾𝐴𝑛 (𝑒𝐼𝜃(𝑖+1) − 2𝑒𝐼𝜃(𝑖) + 𝑒𝐼𝜃(𝑖−1) ) 0.25
Dividing by 𝑒𝐼𝜃𝑖 :
𝐴𝑛+1 − 𝐴𝑛 = 𝛾𝐴𝑛 (𝑒𝐼𝜃 − 2 + 𝑒−𝐼𝜃 )
Dividing by 𝐴𝑛 :
𝐺 = 1 + 𝛾(𝑒𝐼𝜃 − 2 + 𝑒−𝐼𝜃 ) 0.25
Expanding:
𝐺 = 1 + 𝛾(𝐶𝑜𝑠(𝜃) + 𝐼𝑆𝑖𝑛(𝜃) − 2 + 𝐶𝑜𝑠(𝜃) − 𝐼𝑆𝑖𝑛(𝜃))
Simplifying:
𝐺 = 1 + 2𝛾(𝐶𝑜𝑠(𝜃) − 1) 0.25
Then, |𝐺| ≤ 1 → 1 + 2𝛾(𝐶𝑜𝑠(𝜃) − 1) ≤ 1 𝐴𝑁𝐷 1 + 2𝛾(𝐶𝑜𝑠(𝜃) − 1) ≥ −1
The first option is always TRUE since 𝐶𝑜𝑠(𝜃) ≤ 1 0.25
The second option:
1 + 2𝛾(𝐶𝑜𝑠(𝜃) − 1) ≥ −1 → 2𝛾(𝐶𝑜𝑠(𝜃) − 1) ≥ −2 → 𝛾(𝐶𝑜𝑠(𝜃) − 1) ≥ −1
Then 𝛾(1 − 𝐶𝑜𝑠(𝜃)) ≤ 1, so that the worst case scenario (𝐶𝑜𝑠(𝜃) = −1), and for
1 0.25
positive 𝛾, 𝛾(1 − −1) ≤ 1 → 𝛾 ≤ 2

Regarding the Implicit method, only the final part is presented:

1
𝐺=
1 − 2𝛾(𝐶𝑜𝑠(𝜃) − 1)
Thus, 1 ≤ 1 − 2𝛾(𝐶𝑜𝑠(𝜃) − 1) 𝑂𝑅 1 ≥ −1 + 2𝛾(𝐶𝑜𝑠(𝜃) − 1)
1 ≤ 1 − 2𝛾(𝐶𝑜𝑠(𝜃) − 1) → 0 ≤ −2𝛾(𝐶𝑜𝑠(𝜃) − 1)
This is True since the worst case scenario is zero for 𝐶𝑜𝑠(𝜃) = 1.
1 ≥ −1 + 2𝛾(𝐶𝑜𝑠(𝜃) − 1)
2 ≥ 2𝛾(𝐶𝑜𝑠(𝜃) − 1)
1 ≥ 𝛾(𝐶𝑜𝑠(𝜃) − 1)
This is True since the worst case scenario is zero for 𝐶𝑜𝑠(𝜃) = 1.

3
 e) For the sake of simplicity, consider only the steady state version of the transport equation
under analysis, thus:

1) Discretize the first point on the left, where an adiabatic condition was considered. At
least 2 options are available.

Assuming a first order discretization for the thermal flux:

𝑑𝑇 𝑇2 − 𝑇1 0.25
| =0= =0
𝑑𝑥 𝑥=0 𝛥𝑥
The equation for the first point reads as:
𝑇2 − 𝑇1 = 0 0.25

2) Discretize the first point on the left, now i, according to the figure below, where an
adiabatic condition was considered. For that consider the existence of a ghost point,
i-1, but the final expression of the discretization may not have, of course, any reference
to the ghost point. At least 2 options are available. Hint: apply the PDE to point i and
manage to enforce the boundary condition.

The PDE at point i is:

𝑇𝑖+1 − 2𝑇𝑖 + 𝑇𝑖−1 𝑇2 − 2𝑇1 + 𝑇0 0.1
+ 6 = 0 → +6=0
(Δ𝑥)2 (Δ𝑥)2
Relate points i and i-1 with a first order discretization for the thermal flux:
𝑑𝑇 𝑇1 − 𝑇0 0.2
| =0= = 0 → 𝑇0 = 𝑇1
𝑑𝑥 𝑥=0 𝛥𝑥
Then:
𝑇2 − 2𝑇1 + 𝑇0 𝑇2 − 2𝑇1 + 𝑇1 0.2
+6=0→ + 6 → 𝑇2 − 𝑇1 = −6(Δ𝑥)2
(Δ𝑥)2 (Δ𝑥)2

4
 Regarding the Finite Volumes approach:

f) Applying the Finite Volume method to a generic cell P, the process follows as referred in
the next 10 steps.
Fill the green column of the FV process with the label in the green table below.
𝜕𝑇 E
1 = ∇ ∙ (∇𝑇) + 6
𝜕𝑡
𝑡+Δ𝑡 𝑡+Δ𝑡 𝑡+Δ𝑡
𝜕𝑇 C
2 ∫ (∫ 𝑑𝑉) 𝑑𝑡 = ∫ (∫ ∇ ∙ (∇𝑇) 𝑑𝑉) 𝑑𝑡 + ∫ (∫ 6 𝑑𝑉) 𝑑𝑡
𝑡 𝑉 𝜕𝑡 𝑡 𝑉 𝑡 𝑉
𝑡+Δ𝑡 𝑡+Δ𝑡 𝑡+Δ𝑡
𝜕𝑇 A
3 ∫ (∫ 𝑑𝑉) 𝑑𝑡 = ∫ (∫ (∇𝑇) ∙ 𝑛⃗ 𝑑𝑆) 𝑑𝑡 + ∫ (∫ 6 𝑑𝑉) 𝑑𝑡
𝑡 𝑉 𝜕𝑡 𝑡 𝑆 𝑡 𝑉
𝑡+Δ𝑡 𝑡+Δ𝑡 𝑡+Δ𝑡
𝜕𝑇 F
4 Δ𝑥 × 1 ∫ 𝑑𝑡 = ∫ (∫ (∇𝑇) ∙ 𝑛⃗ 𝑑𝑆) 𝑑𝑡 + ∫ 6 Δ𝑥 × 1 𝑑𝑡
𝑡 𝜕𝑡 𝑡 𝑆 𝑡
𝑡+Δ𝑡 𝑡+Δ𝑡 G
5 Δ𝑥(𝑇𝑃𝑛+1 − 𝑇𝑃𝑛 ) =∫ ((∫ (∇𝑇) ∙ 𝑛⃗ 𝑑𝑆)) 𝑑𝑡 + ∫ 6Δ𝑥 𝑑𝑡
𝑡 𝑆 𝑡
𝑛
D
6 Δ𝑥(𝑇𝑃𝑛+1 − 𝑇𝑃𝑛 ) = Δ𝑡 (∫ ((∇𝑇) ∙ 𝑛⃗) 𝑑𝑆) + Δ𝑡(6Δ𝑥 )
𝑆

7 Δ𝑥(𝑇𝑃𝑛+1 − 𝑇𝑃𝑛 ) = Δ𝑡 [ ∑ [(∇𝑇)𝑛𝑖 ∙ ⃗⃗⃗

𝑛𝑖 ]] + 6Δ𝑡Δ𝑥 H
𝑖∈𝑓𝑎𝑐𝑒𝑠
Δ𝑡 B
8 𝑇𝑃𝑛+1 = 𝑇𝑃𝑛 + ∑ [(∇𝑇)𝑛𝑖 ∙ ⃗⃗⃗
𝑛𝑖 ] + 6Δ𝑡
Δ𝑥
𝑖∈𝑓𝑎𝑐𝑒𝑠
Δ𝑡 J
9 𝑇𝑃𝑛+1 = 𝑇𝑃𝑛 + ((∇𝑇)𝑛𝑒 − (∇𝑇)𝑛𝑤 ) + 6Δ𝑡
Δ𝑥
Δ𝑡 𝑇𝐸𝑛 − 𝑇𝑃𝑛 𝑇𝑃𝑛 − 𝑇𝑊𝑛 I
10 𝑇𝑃𝑛+1 = 𝑇𝑃𝑛 + (( )−( )) + 6Δ𝑡
Δ𝑥 Δ𝑥 Δ𝑥

A B C D E F G H I J

Performing the inner

(up to second order)
Original formulation

discretization of the
Integrating in space

Exact integration of

Numerical fluxes
stationary mesh
integration on a
Explicit scheme

diffusive fluxes
Gauss theorem

Second order

Second order
Simplifying

and time

product
LHS

g) Fully explain what was done, and the considered principles, in the steps 4 and 5.

Step 4:
Integration in volume with the trapezoidal rule, thus, up to second order. 0.25
- Assuming the mesh does not vary in time.
- The source term space integration is exact because it is a constant.
Step 5:
Exact mathematical time integration. The integral of the derivative is the value of the 0.25
function taken at the integral limits.

5
 h) Discretize the steady version of the PDE equation in the generic cell P and also in the first
and last Control Volumes.

According to item f), easily one obtains:

∑ [(∇𝑇)𝑖 ∙ ⃗⃗⃗
𝑛𝑖 ] + 6Δ𝑥 = 0 → ((∇𝑇)𝑒 − (∇𝑇)𝑤 ) + 6Δ𝑥 = 0 0.25
𝑖∈𝑓𝑎𝑐𝑒𝑠
For a generic point in the bulk:
𝑇𝐸 − 𝑇𝑃 𝑇𝑃 − 𝑇𝑊 1 0.25
( )−( ) + 6Δ𝑥 = 0 → 𝑇𝑊 − 2𝑇𝑃 + 𝑇𝐸 = −6Δ𝑥 2 → 𝑇𝑊 − 2𝑇𝑃 + 𝑇𝐸 = −
Δ𝑥 Δ𝑥 6
For the first CV on the left:
𝑇𝐸 − 𝑇𝑃 1 0.25
( ) − 0 + 6Δ𝑥 = 0 → −𝑇𝑃 + 𝑇𝐸 = −6Δ𝑥 2 → −𝑇𝑃 + 𝑇𝐸 = −
Δ𝑥 6
For the last CV on the right:
100 − 𝑇𝑃 𝑇𝑃 − 𝑇𝑊 1 0.25
( )−( ) + 6Δ𝑥 = 0 → 𝑇𝑊 − 3𝑇𝑃 = −6Δ𝑥 2 − 200 → 𝑇𝑊 − 3𝑇𝑃 = − − 200
Δ𝑥 Δ𝑥 6
2

i) Regarding the previous discretization, fill ALL the entries in the system of equations
below:

1
-1 1 𝑻𝟏 −
6
1
1 -2 1 𝑻𝟐 −
6
1
1 -2 1 𝑻𝟑 = −
6
1
1 -2 1 𝑻𝟒 −
6
1
1 -2 1 𝑻𝟓 −
6
1
1 -3 𝑻𝟔 − − 200
6

0.5

6
Problem 2 (2.0 Marks)
In the context of the Finite Volume
Method, the first order upwind scheme
for the convective term, together, or not,
with the first order explicit Euler for time
discretization, present some specific
properties that may be good or bad,
depending on the context and the
expectations.
In the following you are asked to express
the pros and cons of this (these) first
order scheme (schemes).
Please address, at least, 6 of the following ‘keywords’ or ideas:
- boundedness - monotonicity
- dissipation - diagonal dominancy
- dispersion - order of accuracy
- exact solution for the advection - numerical diffusion
equation under specific conditions - Riemann solver
- flux limiter - accuracy and robustness

7
Each correct and justified expression …………………………………………………………… -> 0.33
Some examples in the following:

FOU ensures boundedness GOOD

FOU + EE is dissipative and dispersive, but at a simple glance, dissipation is the seen
picture BAD
FOU is a monotone scheme also according to Godunov theorem GOOD
FOU + EE are lower order schemes in space and time needing (typically) high spatial
and temporal resolution for accurate results BAD
FOU alone or + EE generates a significant amount of numerical diffusion, due to its
dissipative nature BAD
FOU for steady problems allows for a diagonally dominant system of equations GOOD
FOU + EE displays an exact solution for the 1D linear advection equation with uniform
grid, periodic boundary conditions and unit Courant. Good for debugging GOOD
FOU alone or + EE allows for simplicity in the coding, and are efficient and robust (at a
cost of accuracy) GOOD
FOU is a flux limiter itself and an important part of any flux limiter to effectively deal
with local extrema GOOD
FOU even as a part of a flux limiter, limits the local order of accuracy to one close to local
extrema BAD
FOU + EE is an exact Riemann solver for the linear advection equation following the
Godunov’s method GOOD
FOU + EE is an exact Riemann solver for the linear advection equation but too much
dissipative indeed. Even with the right physical modelling, one needs a very fine mesh BAD
for a good and accurate shock resolution

8
 Problem 3 (0.5+0.5+1.0+1.0 = 3.0 Marks)
Consider the duct network bellow, for an incompressible fluid, feeding an industrial meat facility
where each compartment is kept at the same constant pressure 𝑃 = 0.
The duct network comprises 5 pipes to be calculated with momentum equation given by:
−Δ𝑃 = 𝛽𝑉̇
̇
where 𝑃 is the pressure, 𝑉 is the volumetric flow rate and β = 5.
The continuity equation, to be solve in the unknown black nodes, can be taken as:
∑𝑉 ⃗ ̇ ∙ ⃗⃗⃗
𝑛𝑖 = 0

a) A staggered mesh to the solution of the problem is shown below. Justify the relevance of
considering a staggered mesh. Also justify this mesh for this problem, and which equation
(and why?) is to be considered in the green, yellow and purple CV’s.
A staggered mesh avoids spurious pressure oscillations and, in the same way, reduces the 0.2
number of needed interpolations to the faces
The mesh is suitable for this problem since each CV is ‘allocated’ for one unknown and the CV’s
in the boundary of the system have a volumetric flow rate or pressure boundary condition 0.2
The purple CV is for pressure, thus a scalar, and is to the discretization of the continuity equation.
Green and yellow CV’s are for the components of ‘velocity’ and thus for each component of the
momentum equation. In this simple 1D problem only one component of momentum equation 0.1
exists. The present distinction between green and yellow CV’s is just to mimic a 2D domain.

9
 b) Conclude, with full detail, that the volumetric flow rate correction is related to the
pressure correction in the following way:
Δ𝑃′
𝑉̇ ′ = −
𝛽
∗ ′ ̇
Assuming 𝑃 = 𝑃 + 𝑃 and 𝑉 = 𝑉 + 𝑉 ̇ ∗ ̇ ′ 0.1
The exact equation is −Δ𝑃 = 𝛽𝑉̇
But the computed equation, at each stage, and which is ‘solved’ is −Δ𝑃∗ = 𝛽𝑉̇ ∗ 0.2
Subtracting it, and considering the linearity of the equation and operators:
−Δ𝑃′ 0.2
−Δ(𝑃 − 𝑃∗ ) = 𝛽(𝑉̇ − 𝑉̇ ∗ ) → −Δ𝑃′ = 𝛽𝑉̇ ′ → 𝑉̇ ′ =
𝛽

c) Derive, with full detail, the ‘SIMPLE’ algorithm to obtain the following system of equations
for the pressure correction.
Solving continuity in the purple CV’s for Knowing that 𝑉̇ = 𝑉̇ ∗ + 𝑉̇ ′ :
the (exact) 𝑉𝛼̇ :
𝑉̇2 + 𝑉𝑆̇ − 𝑉1̇ = 0 𝑉̇2′ + 𝑉𝑆̇ ′ − 𝑉1̇ ′ = −(𝑉̇2∗ + 𝑉𝑆̇ ∗ − 𝑉1̇ ) 0.3
𝑉̇3 + 𝑉̇𝑃 − 𝑉̇2 = 0 𝑉̇3′ + 𝑉̇𝑃′ − 𝑉̇2′ = −(𝑉̇3∗ + 𝑉̇𝑃∗ − 𝑉̇2∗ )
𝑉̇𝐶 − 𝑉̇3 = 0 𝑉̇𝐶′ − 𝑉̇3′ = −(𝑉̇𝐶∗ − 𝑉̇3∗ )
The momentum (green and yellow CV’s with the arrows) equations are, in fact, the
‘velocity’ correction equations, as known, and read as:
𝑃𝑆′ − 𝑃1′ = −𝛽𝑉𝑆̇ ′
𝑃𝑃′ − 𝑃2′ = −𝛽𝑉̇𝑃′
𝑃𝐶′ − 𝑃3′ = −𝛽𝑉̇𝐶′ 0.2
𝑃2′ − 𝑃1′ = −𝛽𝑉̇2′
𝑃3′ − 𝑃2′ = −𝛽𝑉̇3′
Δ𝑃′
From the equation for flow correction 𝑉̇ ′ = − obtain:
𝛽
(𝑃1′ − 𝑃2′ ) + (𝑃1′ − 𝑃𝑆′ ) − 𝑉1̇ ′ = −𝛽(𝑉̇2∗ + 𝑉𝑆̇ ∗ − 𝑉1̇ ) 0.3
(𝑃2′ − 𝑃3′ ) + (𝑃2′ − 𝑃𝑃′ ) − (𝑃1′ − 𝑃2′ ) = −𝛽(𝑉̇3∗ + 𝑉̇𝑃∗ − 𝑉̇2∗ )
(𝑃3′ − 𝑃𝐶′ ) − (𝑃2′ − 𝑃3′ ) = −𝛽(𝑉̇𝐶∗ − 𝑉̇3∗ )
Applying the boundary conditions 𝑃𝑆′ = 𝑃𝑃′ = 𝑃𝐶′ = 𝑉1̇ ′ = 0 then,
(𝑃1′ − 𝑃2′ ) + (𝑃1′ ) = −𝛽(𝑉̇2∗ + 𝑉𝑆̇ ∗ − 𝑉1̇ )
(𝑃2 − 𝑃3′ ) + (𝑃2′ ) − (𝑃1′ − 𝑃2′ ) = −𝛽(𝑉̇3∗ + 𝑉̇𝑃∗ − 𝑉̇2∗ )
′ 0.1
(𝑃3′ ) − (𝑃2′ − 𝑃3′ ) = −𝛽(𝑉̇𝐶∗ − 𝑉̇3∗ )
Simplifying,
2𝑃1′ − 𝑃2′ = −𝛽(𝑉̇2∗ + 𝑉𝑆̇ ∗ − 𝑉1̇ )
−𝑃1′ + 3𝑃2′ − 𝑃3′ = −𝛽(𝑉̇3∗ + 𝑉̇𝑃∗ − 𝑉̇2∗ ) 0.1
−𝑃2′ + 2𝑃3′ = −𝛽(𝑉̇𝐶∗ − 𝑉̇3∗ )

𝟐 −𝟏 𝟎 𝑷′𝟏 𝜷(𝟏𝟎𝟎 − 𝑽̇∗𝟐 − 𝑽̇∗𝑺 )

−𝟏 𝟑 −𝟏 𝑷′𝟐
= 𝜷(𝑽̇∗𝟐 − 𝑽̇∗𝟑 − 𝑽̇∗𝑷 )

𝟎 −𝟏 𝟐 𝑷′𝟑 𝜷(𝑽̇∗𝟑 − 𝑽̇∗𝑪 )

10
 d) Conduct one iteration of the process, justifying why one single iteration is enough. Thus,
define an initial guess for the pressure, estimate the velocity field and present the
pressure correction system of equations to be solved.

Only one iteration is enough because of 2 reasons:

- the equations are linear 0.2
- the velocity correction on a CV does not depend on the velocity corrections on 0.1
the neighborhood, thus no simplifications are needed when deriving the SIMPLE

Let’s conduct one iteration of the process. Here is just presented a possible way…
Initial guess for the pressure:
𝑃1∗ = 𝑃2∗ = 𝑃3∗ = 0
and remember that, 0.1
𝑃𝑆 = 𝑃𝑃 = 𝑃𝐶 = 0
Estimation of the ‘velocity field’:
𝑃𝑆 − 𝑃1∗ = −𝛽𝑉𝑆̇ ∗
𝑃𝑃 − 𝑃2∗ = −𝛽𝑉̇𝑃∗
𝑃𝐶 − 𝑃3∗ = −𝛽𝑉̇𝐶∗
𝑃2∗ − 𝑃1∗ = −𝛽𝑉̇2∗
𝑃3∗ − 𝑃2∗ = −𝛽𝑉̇3∗
Thus: 0.3
𝑉𝑆̇ ∗ = 0
𝑉̇𝑃∗ = 0
𝑉̇𝐶∗ = 0
𝑉̇2∗ = 0
𝑉̇3∗ = 0
The pressure correction system below

𝟐 −𝟏 𝟎 𝑷′𝟏 𝜷(𝟏𝟎𝟎 − 𝑽̇∗𝟐 − 𝑽̇∗𝑺 )

−𝟏 𝟑 −𝟏 𝑷′𝟐
= 𝜷(𝑽̇∗𝟐 − 𝑽̇∗𝟑 − 𝑽̇∗𝑷 )

𝟎 −𝟏 𝟐 𝑷′𝟑 𝜷(𝑽̇∗𝟑 − 𝑽̇∗𝑪 )

should look like:

𝟐 −𝟏 𝟎 𝑷′𝟏 𝟓𝟎𝟎

−𝟏 𝟑 −𝟏 𝑷′𝟐
= 𝟎
0.3

𝟎 −𝟏 𝟐 𝑷′𝟑 𝟎

11
12
Problem 4 (2.0+1.0+0.5 = 3.5 Marks)

Consider the unstructured mesh on the right, where a

detail is presented below with the pertinent points on
the surrounding properly labeled.
A local Coordinate system is also represented below.
All sides of the pentagons have a length of 1.

a) Considering the transport equation 0 = ∇ ∙ (∇(𝜙)), apply the Finite Volume method in a
simplified way, taking in mind the cell P. Discretize the transport equation in the control
volume P. Comment the expected order of accuracy of the diffusive term, and the need, or
not, to consider non-orthogonal corrections.
Applying the FVM:
∫ ∇ ∙ (∇𝑇) 𝑑𝑉 = 0 → ∫ (∇𝑇) ∙ 𝑛⃗ 𝑑𝑆 = 0 → ∑ [(∇𝑇) ∙ ⃗⃗⃗ 𝑛 ]𝐴 = 0 0.5
𝑖 𝑖 𝑖
𝑉 𝑆 𝑖∈𝑓𝑎𝑐𝑒𝑠
For this particular situation with no warp angle:
𝑑𝑇
∑ 𝐴 =0
𝑛𝑖 𝑖
𝑑⃗⃗⃗
𝑖∈𝑓𝑎𝑐𝑒𝑠
Thus, discretizing,
𝑇𝑁𝑒𝑖𝑔ℎ𝑏𝑜𝑟 − 𝑇𝑃 0.3
∑ ×1=0
𝑑𝑃 𝑡𝑜 𝑁𝑒𝑖𝑔ℎ𝑏𝑜𝑟
𝑖∈𝑓𝑎𝑐𝑒𝑠
1
𝑑𝑃 𝑡𝑜 𝑁𝑒𝑖𝑔ℎ𝑏𝑜𝑟 = 2 × × 𝑇𝑔(54°) = 𝑇𝑔(54°) = 1.376 0.2
2
For CV P:
𝑇𝐴 − 𝑇𝑃 𝑇𝐵 − 𝑇𝑃 𝑇𝐶 − 𝑇𝑃 𝑇𝐷 − 𝑇𝑃 𝑇𝐸 − 𝑇𝑃
+ =0
𝑑𝑃 𝑡𝑜 𝑁𝑒𝑖𝑔ℎ𝑏𝑜𝑟 𝑑𝑃 𝑡𝑜 𝑁𝑒𝑖𝑔ℎ𝑏𝑜𝑟 𝑑𝑃 𝑡𝑜 𝑁𝑒𝑖𝑔ℎ𝑏𝑜𝑟 𝑑𝑃 𝑡𝑜 𝑁𝑒𝑖𝑔ℎ𝑏𝑜𝑟 𝑑𝑃 𝑡𝑜 𝑁𝑒𝑖𝑔ℎ𝑏𝑜𝑟
0.5
𝑇𝐴 + 𝑇𝐵 + 𝑇𝐶 + 𝑇𝐷 + 𝑇𝐸 − 5𝑇𝑃 = 0
The expected order of accuracy is 2. The diffusive flux, with no warp angle and no
skewness, is in fact a pure central differences scheme. 0.5
The corrections are related to the warp angle. No warp angle means no need to non-
orthogonal corrections.

13
 b) Consider the general expressions below for the components of the Cell ‘average’ Gradient
for cell P:

̅̅̅̅̅̅̅̅
∇(𝜙)𝑃 |𝑥 = 𝛼1 𝜙𝑃 + 𝛼2 𝜙𝐴 + 𝛼3 𝜙𝐵 + 𝛼4 𝜙𝐶 + 𝛼5 𝜙𝐷 + 𝛼6 𝜙𝐸

̅̅̅̅̅̅̅̅
∇(𝜙)𝑃 |𝑦 = 𝛽1 𝜙𝑃 + 𝛽2 𝜙𝐴 + 𝛽3 𝜙𝐵 + 𝛽4 𝜙𝐶 + 𝛽5 𝜙𝐷 + 𝛽6 𝜙𝐸

Say, briefly justifying, with no need of calculations, which 𝛼 or 𝛽 above are zero. Comment
the expected order of accuracy for the Cell Gradient on cell P.

In both x and y components the coefficient of 𝜙𝑃 is zero by symmetry.

Since the weight of 𝜙𝑃 in each face is the same, 0.5, then it will cut in the end because
the ∑𝑓=𝑛𝑏(𝑃) 𝑺𝒇 for each component (x and y) is zero by definition. 0.4
Thus 𝛼1 = 0 and 𝛽1 = 0.
In the x component 𝛼4 = 0 because there’s no x normal component in the face the 0.4
receives the contribution of 𝜙𝐶 .
The discretization of the cell gradient is done by central interpolation, a second order
scheme at prior, namely in uniform arrangements. Since the interpolation falls into the 0.2
face’s centroid, then the integration of the ‘flux’ will be second order. Overall the
expected order of accuracy is 2.

Formulae
1 1
̅̅̅̅̅̅
∇𝜙 𝑃 = ∫ ∇𝜙 𝑑𝑉 ̅̅̅̅̅̅
∇𝜙 𝑃 = ∫ 𝜙 𝑑𝑺
𝑉𝑃 𝑉𝑃 𝑉𝑃 𝜕𝑉𝑃
1
̅̅̅̅̅̅
∇𝜙 𝑃 𝑉𝑃 = ∑ ∫ 𝜙 𝑑𝑺 ̅̅̅̅̅̅
∇𝜙 𝑃 = ∑ ̅̅̅̅
𝜙𝑓 𝑺𝒇
𝑓𝑎𝑐𝑒 𝑉𝑃
𝜕𝑉𝑃 𝑓=𝑛𝑏(𝑃)

c) Regarding cell A, comment the expected order of accuracy for the Cell Gradient. Also
comment the need of non-orthogonal corrections for the diffusive flux.

For the cell gradient, the contribution of neighbor G (for example) does not fall into the
centroid of the face dividing cells A and G. It’s a skewness problem that limit the order 0.3
of accuracy of the cell gradient to one.
In the same face, is evident the existence of a warp angle, thus non-orthogonal 0.2
correction will be expected to avoid the scheme do not be consistent.

14
Problem 5 (20 x 0.2 = 4.0 Marks) Wrong answers discount 50% (-0.1 Mark)

Indicate if the statements below are True or False.

T F
1 The Burger’s equation is hyperbolic even for small velocities, close to zero. x
When discretizing the linear one-dimensional convection equation with an
2 explicit scheme, the application of a flux limiter makes the numerical scheme to x
be non-linear.
3 Linear 1D advection equation only displays contact discontinuities, while 1D
x
Burger’s equation may display shocks and non-isentropic expansion waves.
Godunov method employs a piecewise-constant reconstruction and evolves
4 according to the exact Riemann problem solution at interfaces of the control
x
volumes and the time step is controlled so that the face Riemann problems are
not allowed to interact.
5 The SOR method with an over-relaxation factor 𝜆 > 2 can speed up the
x
convergence by a factor higher than 1.
6 For the linear one-dimensional advection equation, the analytical phase angle is
x
unitary.
7 PWIM method (Rhie and Chow) calculates the corrections for the velocity at the
x
faces of the control volumes.
8 If a scheme is stable and consistent, then the scheme will be convergent. x
9 The outflow boundary conditions needed in the simulation of 3D Euler
x
compressible flows are not the same for subsonic and supersonic flow.
If you perform some numerical experiments computing discontinuous solutions
10 using Godunov’s method, you observe that Godunov’s method adds some x
artificial smearing to the solution.
11 Godunov’s method is an exact Riemann’s solver and thus, in general, we may have
x
exact solutions of discontinuous fields.
12 When using the Grid Convergence Index (GCI) method, at least three meshes
x
must be used, hopefully out of the asymptotic range of convergence.
A polygonal mesh may be the dual mesh of a triangular mesh only if Delaunay
x
13 triangulation was employed.
The Gauss-Seidel method, for the numerical solution of linear systems of
14 equations, may be used for the convergence process of non-linear systems of x
equations.
15 In 2D, a set of 3251 dispersed points, thus not collinear, always generate a
x
Delaunay triangulation.
In unstructured grids, we may consider solely the cell gradients to compute the
16 diffusive fluxes, with no non-orthogonal corrections. However numerical stability x
problems may occur.
17 Five possible wave patterns exist in the solution of the Riemann problem for the
x
inviscid 1D Burgers equation, when evaluating the Godunov flux.
18 Coupled solvers are more suitable to compressible flows than segregated solvers. x
19 The SIMPLE algorithm is a Projection Method with orthogonal projection of the
x
fields onto the solenoidal space.
20 The SIMPLEC algorithm does not consider the contribution of the neighbors to
x
the velocity correction.

15